BEGIN:VCALENDAR
VERSION:2.0
CALSCALE:GREGORIAN
PRODID:UW-Madison-Physics-Events
BEGIN:VEVENT
SEQUENCE:1
UID:UW-Physics-Event-6457
DTSTART:20210602T160000Z
DTEND:20210602T171500Z
DTSTAMP:20260414T220433Z
LAST-MODIFIED:20210522T215306Z
LOCATION:Online Seminar: Please sign up for our mailing list at www.ph
 ysicsmeetsml.org for zoom link
SUMMARY:Deep Neural Networks for Ab Initio Quantum Chemistry\, Physics
  ∩ ML Seminar\, David Pfau\, Deepmind
DESCRIPTION:In this talk\, I will present work on how ideas from the m
 achine learning community can give back to computational physics\, in 
 particular deep neural networks and approximate natural gradient desce
 nt. I will present a novel deep neural network architecture\, the Ferm
 ionic Neural Network (FermiNet)\, which can be used as an expressive c
 lass of approximate solutions (Ansätze) to the Schrödinger equation 
 for many-electron systems. We optimize the FermiNet by Kronecker-Facto
 rized Approximate Curvature (KFAC)\, an approximation to natural gradi
 ent descent which can also be used to approximate stochastic reconfigu
 ration. This makes it possible to scale stochastic reconfiguration to 
 Ansätze with large numbers of parameters. We show that the FermiNet i
 s able to achieve much higher accuracy than standard variational QMC A
 nsätze like the Slater-Jastrow-backflow ansatz\, and can exceed the a
 ccuracy of coupled cluster methods like CCSD(T) on bond-breaking syste
 ms like the transition of bicyclobutane to butadiene. This shows that 
 deep neural networks can be used to greatly improve the accuracy of va
 riational QMC\, to the point where it is competitive with other state-
 of-the-art ab initio methods.
URL:https://www.physics.wisc.edu/events/?id=6457
END:VEVENT
END:VCALENDAR